Jure Leskovec (Computer Science)
Light refreshments will be provided.
This week Jure Leskovec joins the Network Forum to discuss two frontiers in network analysis: cascades and communities. We will divide the 90 minute Forum into two sessions and discuss each topic for 45 minutes. Jure will begin each session by briefly presenting the first paper listed under each topic below. Participants will come prepared to discuss these two papers which have been attached for convenience. The second paper listed under each topic is supplemental reading for participants wishing to discuss that topic with Jure in greater detail.
Topic 1: Cascades
- Can Cascades be Predicted? by J. Cheng, L. Adamic, A. Dow, J. Kleinberg, J. Leskovec. ACM International Conference on World Wide Web (WWW), 2014.
- What’s in a name? Understanding the Interplay between Titles, Content, and Communities in Social Media by H. Lakkaraju, J. McAuley, J. Leskovec. AAAI International Conference on Weblogs and Social Media (ICWSM), 2013.
Cascades are a useful concept in the study of diffusion and contagion processes. The first paper presents the finding that cascades can be predicted by temporal and structural features. The authors conduct an experiment with photo-sharing in Facebook and predict cascades' eventual shape and size. The second and supplemental paper for this session is a study of content resubmission in Reddit. An interesting combination of communities, language, and cascades, the authors find that the language used in content titles predicts variation in resubmission cascades across independent Reddit communities.
Topic 2: Communities
- First paper (draft/unpublished): http://stanford.io/1k7rQ6Z
- Detecting Cohesive and 2-mode Communities in Directed and Undirected Networks by J. Yang, J. McAuley, J. Leskovec. ACM International Conference on Web Search and Data Mining (WSDM), 2014.
The study of communities is critical for those who wish to uncover the functional modules and systems underlying their blob of nodes and edges. The first paper forwards a not-yet-published paradigm that "dense network cores occur as the convolution of many overlapping communities". As the title suggests, deconvolution of these dense cores reveals the core-periphery organization of these communities. The second and supplemental paper for this session extends earlier community detection methods to bipartite and directed networks.
Jure Leskovec is an assistant professor of Computer Science at Stanford University. His research focuses on mining and modeling large social and information networks, their evolution, and diffusion of information and influence over them. Problems he investigates are motivated by large scale data, the Web and on-line media.